Hybrid Strategies Towards Safe “Self-Aware” Superintelligent Systems